The Research And Implementation Of Embedded Virtual Roaming Scene’s Dynamical Generation

In recent years, the research and implementation development on virtual reality, which isemerging technology and related to computer graphics, multimedia and human-machineinterface technology, is very rapid. It has been widely used in so many fields, such as education,entertainment, architectural design and so on. Along with the increase of the standard of living,people’s traveling frequency is much higher then before. Thus, the demand for the navigationsystem shows an increasing trend too. Because the embedded navigation devices have manyadvantages, their market prospect is pticularly considerable.As an essential function of navigation equipment, electronic maps have already basiclyprosuctized. However, most of the navigation equipments are still using traditionaltwo-demensional map. Although in this way, the map can deliver the query results via specificsymbols, but it can not express the intuitive and vivid concepts. The improvement of embeddeddevices’ performance makes it possible for the three-dimensional graphics expand to them.Three-dimensional electronic map can not only provide basic location information, but also beable to give intuitive viewing experiences during the process of navigation. As a developmentderiction of electronic map, it has been got more and more people’s concern. This is theimplementation of virtual reality in the navigation field.In the three-dimensional electronic map function implementation process, one key issue isabout frame rate guarantee. During roaming scene, if the system needs to take several secondsto respond each UI event, such as moving or ratating the camera, which is human-computerinterface provided by the system, this system is totally unacceptable. How to improve the speed of the virtual scene grneation during the process of real-time roaming on the embedded deviceshas become a hot research issue.This paper describes a system which uses OpenGL ES provided natively by Androidplatform to render three-dimensional models and display in the main App window. In the aspectof scene generation, it gets helps from commonly used modeling sofeware named3Ds max. Inresponse to those issues mentioned before, the speed improvement algorithms such as the scenemanagement, LOD multi-resolution models, scene objects fast scheduling strategy has beenstudied. Focusing on the scene management, LOD models generation and scene objectsscheduling strategy, this paper presents some improvements:(1) The three-dimensional digital map contains a number of separate scenes. This paperproposes an adjustable Octree division method to deal with complex scenes problems. Theadjustment mechanism is based on the distribution of vertices of the meshes, which can makevertices be concentrated as much as possible in a few numbers of nodes in the scene. In thisway, the memory overhead can be reduced and the efficiency of model partiontioning algorithmcan also be improved. In order to improve the efficiency of system startup, scene objects will beserialized to external storage, which can be used directly next time.(2) In the part of the mesh simplification, this paper proposes the simplification approachusing weights based on the curvature of the triangle combining with the quadric error metricbased on distance, which can simplify the scene models to several different levels. These resultscan be stored and selected dynamically based on the distance factor. This approach uses eachscene management object as a unit. Because the mesh topology structure can be changed solong as models simplification happens, the new vertex location will be the key issue if it isnecessary to maintain visual features.This paper proposes a vertex selection mode by which thevertices can be reused. Besides, this mode combines with punishment mechanism so that it canmaintain models’ visual fertures as much as possible.(3) In the part of scene scheduling, the scheduling mechanism based on Morton code ispresented. Whenever users change the position and direction of the roaming object, it canquickly find out the currently visible nodes set in the scene by bit computing. As the result, theset will be added into rendering queue to render and display later.(4) In the collision detection part, using parallel traversal model AABB (Axis-Aligned Bounding Boxes) tree approach, this system can improve the detection efficiency.In this paper, these mentioned accelerative algorithms are implemented by using OpenGLES on the Android OS platform. Experiments show that these methods can reduce the memoryoverhead of the system while reducing the computational complexity when preprocessing;scenario scheduling strategy increases the flexibility of scheduling while the efficiency could bealso guaranteed. Totally, these are reflected in the improvement of FPS.Finally, the consequences of the research are briefly summarized, and then the paperdescribes the future research direction which is likely to be continued.